Correlation between Coronavirus (COVID-19) Vaccinations and Excess Mortality in the OWID Dataset

import warnings
warnings.simplefilter(action='ignore', category=FutureWarning)
import pandas as pd
import numpy as np
import plotly.express as px
import plotly.figure_factory as ff

Get and Process the Data

We get the data from OWID and interpolate/upsample the missing values weekly.

d=pd.read_excel('https://covid.ourworldindata.org/data/owid-covid-data.xlsx')

Filter Out Countries With Insufficient Data

If there is no single row with both Vaccinations and Excess Mortality data we have nothing to work with. So we filter such countries out completely.

country_corr=d.groupby('iso_code')[
    ['excess_mortality','total_vaccinations_per_hundred']
                                  ].corr().unstack().iloc[:,1].reset_index()

country_corr.columns=['iso_code','correlation']

countries_worth_looking_at=list(
    country_corr[~country_corr['correlation'].isna()]
    .sort_values(
        'correlation'
        ,ascending=False
    )['iso_code']
)

country_corr_dict=dict(zip(country_corr.iso_code, country_corr.iso_code + country_corr.correlation.apply( " {0:.2f}".format) ))

dfc = d[(d['iso_code'].isin(countries_worth_looking_at))].copy()

Interpolate The Data By Backfilling

Some countries have the data reported monthly while others have missing data points. We resample everything to the weekly grane and interpolate via backfilling. This way the vaccination time series before the start of reporting are backfilled with a static value (first value in time serie), mostly close to zero. This makes the year 2020 kind of a synthetic control in regard to the vaccination treatment.

# resample needs datetime index
dfc['datetime'] = pd.to_datetime(dfc['date'])
dfc.index = dfc['datetime']
del dfc['datetime']

# we use the mean() function to fill the missing values with the NAs
dfi=dfc.groupby('iso_code').resample('W').mean().reset_index()
dfi.index = pd.to_datetime(dfi['datetime'])
del dfi['datetime']

# interpolation within countries
for i in range(len(dfi.iso_code.unique())):
    mask = dfi.loc[:,'iso_code']==dfi.iso_code.unique()[i]
    dfi[mask]=dfi[mask].interpolate(method='bfill')
dfi.reset_index(inplace=True)

Overall Correlation between Vaccinations and Excess Mortality

print('Correlation total_vaccinations_per_hundred and excess_mortality ='
      ,'\x1b[1;30;30m{:.2f}\x1b[m'.format(
    dfi[['excess_mortality', 'total_vaccinations_per_hundred']].corr().unstack()[1])) 

Correlation total_vaccinations_per_hundred and excess_mortality = -0.03

At the moment of writing the correlation is close to zero (-0.04). Let's de-average the overal value by calculating the correlation per country.

De-averaging Of The Correlation By Country

We chart correlation between excess_mortality and total_vaccinations_per_hundred (and other vaccination metrics) per country to make sure the data is sane.

The Choropleth Charts give an idea how the correlation is distibuted over the globe. May be there is some role the seasonality plays in the resulting in-country corellation.

The Histograms show that the in-country correlation is kinda evenly distributed over the whole [-1:1] interval and there are no visible ouliers.

for metric in ['total_vaccinations_per_hundred'
               , 'people_vaccinated_per_hundred'
               , 'people_fully_vaccinated_per_hundred'
               , 'total_boosters_per_hundred'
               , 'new_cases_per_million' ]:
    
    country_corr=dfc.groupby('iso_code')[
        ['excess_mortality', metric]
                   ].corr().unstack().iloc[:,1].reset_index()
    
    colname = 'correlation excess_mortality and ' + metric

    country_corr.columns=['iso_code', colname]
    
    fig = px.choropleth (
        country_corr,
        locationmode = 'ISO-3',
        locations = 'iso_code',
        color = colname,
    )
    fig.update_traces(
        showlegend=False
        , selector=dict(type='choropleth'))
    fig.update_layout(
        width=2048,
        height=800,
        title_text=colname,
        geo=dict( showcoastlines=False,),
        coloraxis_colorbar=dict( title='Scale, Correlation',),
        )
    fig.show('png')   
    
    fig=ff.create_distplot(
        [country_corr.dropna()[colname]]
        , [colname]
        , bin_size=.2
        , histnorm = 'probability'
    )
    fig.update_layout(
        title_text='Country distribution over the correlation [-1:1] and kernel density estimation'
        , xaxis_title="Countries over Correlation per Country"
        , yaxis_title="Probabililty and Kernel Density Estimation"
        , legend_x=0 
    )
    fig.show()

Interpolated Data For Visual Control

# let's add a helper column for facet titles
dfi['cntr']=dfi['iso_code'].replace(country_corr_dict, inplace=False)

fig = px.line(dfi,    
              x='datetime',
              y=['total_vaccinations_per_hundred','excess_mortality'],
              facet_col='cntr', 
              category_orders={'iso_code':sorted(countries_worth_looking_at)},
              facet_col_wrap=4,
              facet_row_spacing=0.01,
              height=6000, #width=800, # change to fit your output media 
              labels={
                  "value": "per hundred / percent",
                  "variable": "",
              },
              title="Country Charts and Corr, Time Series Backfilled Weekly, Jan 2020 - Jan 2022",              
             )
           
fig.update_layout(
    legend=dict(
        yanchor="top",
        y=0.999,
        xanchor="left",
        x=0.01
    ),
    xaxis_title="Weekly Data Jan 2020 - Jan 2022"             
)

fig.update_yaxes(range=[-50, 200])

fig.show() # we can't use 'png' here in JupyterLab as the charts get scrambled, a plotly bug?

Obligatory XKCD parabole

Permanent link to this comic: https://xkcd.com/925/

Vaccine is present/absent

Correlation between excess_mortality and total_vaccinations_per_hundred in countries where a vaccine is present/absent

For each vaccine we split all countries into two groups as per vaccine presence and calculate the correlation within those two group. Obviously there are countries with multiple vaccine present. And some vaccine may only come together in all countries. Yet the results are interesting and non-trivial.

v=pd.read_csv('https://github.com/owid/covid-19-data/raw/master/public/data/vaccinations/locations.csv', sep=',', quotechar='"'
              , header=0, usecols=['iso_code', 'vaccines'], dtype={'iso_code':'str','vaccines':'str'} )

from sklearn.preprocessing import MultiLabelBinarizer 
# convert to contain lists instead of plain text 
# TODO: get rid of leading spaces or whatever blanks they are instead of dropping all spaces.
v['vaccines']=v['vaccines'].str.replace(r'[^a-zA-Z,]+',r'').str.split(',') 

# break into indicator variables 
mlb = MultiLabelBinarizer() 
indicators = pd.DataFrame(mlb.fit_transform(v['vaccines']), 
                          columns=mlb.classes_, index=v.index)
vac_list=list(mlb.classes_)
vac_list

['Abdala',
 'COVIranBarekat',
 'CanSino',
 'Covaxin',
 'EpiVacCorona',
 'JohnsonJohnson',
 'Medigen',
 'Moderna',
 'Novavax',
 'OxfordAstraZeneca',
 'PfizerBioNTech',
 'QazVac',
 'SinopharmBeijing',
 'SinopharmWuhan',
 'Sinovac',
 'Soberana',
 'SputnikLight',
 'SputnikV',
 'Turkovac',
 'ZF']

v = pd.concat([v.reset_index(drop=True), 
                indicators.reset_index(drop=True)], axis=1) 

v.set_index('iso_code', drop=True, inplace=True)

dfi=dfi.join(v, on='iso_code')

def get_vac_corr(vac):
        vac_corr=dfi.groupby(vac)['excess_mortality','total_vaccinations_per_hundred'].corr() 
        vac_corr1 = vac_corr.stack().reset_index()
        vac_corr1.columns = ['vaccine_presence_ind','b','c','correlation']
        vac_corr1.vaccine_presence_ind.replace([0,1],['absent','present'], inplace=True)
        vac_corr1['vaccine_name']=vac
        corr_per_vac = vac_corr1.loc[ 
            vac_corr1.b.isin(['excess_mortality']) & vac_corr1.c.isin(['total_vaccinations_per_hundred'])
            , ['vaccine_name','vaccine_presence_ind','correlation'] 
        ] 
        return corr_per_vac   
    
vac_corrs = pd.concat(  get_vac_corr(vac) for vac in vac_list).reset_index() 
del vac_corrs['index']
print(vac_corrs)

         vaccine_name vaccine_presence_ind  correlation
0              Abdala               absent    -0.034223
1      COVIranBarekat               absent    -0.031186
2      COVIranBarekat              present    -0.194205
3             CanSino               absent    -0.029848
4             CanSino              present    -0.104211
5             Covaxin               absent    -0.027065
6             Covaxin              present    -0.220093
7        EpiVacCorona               absent    -0.036802
8        EpiVacCorona              present     0.652400
9      JohnsonJohnson               absent    -0.041634
10     JohnsonJohnson              present    -0.020981
11            Medigen               absent    -0.036992
12            Medigen              present     0.323308
13            Moderna               absent    -0.022750
14            Moderna              present    -0.011094
15            Novavax               absent    -0.034223
16  OxfordAstraZeneca               absent     0.027271
17  OxfordAstraZeneca              present    -0.046115
18     PfizerBioNTech               absent     0.111482
19     PfizerBioNTech              present    -0.032134
20             QazVac               absent    -0.037213
21             QazVac              present     0.420184
22   SinopharmBeijing               absent    -0.028825
23   SinopharmBeijing              present     0.012047
24     SinopharmWuhan               absent    -0.034223
25            Sinovac               absent    -0.020833
26            Sinovac              present    -0.040698
27           Soberana               absent    -0.031186
28           Soberana              present    -0.194205
29       SputnikLight               absent    -0.036788
30       SputnikLight              present     0.147485
31           SputnikV               absent    -0.014283
32           SputnikV              present     0.014178
33           Turkovac               absent    -0.034223
34                 ZF               absent    -0.034216
35                 ZF              present    -0.101346

fig = px.bar(vac_corrs
             ,x='vaccine_name'
             ,y='correlation'
             ,color='vaccine_presence_ind'
             ,color_discrete_sequence=["grey","blue",]
             ,barmode="group"
             ,title="""Correlation between excess_mortality and total_vaccinations_per_hundred 
<br>in countries grouped by vaccine presence/absence."""
             , height=600, width=1000, # change to fit your output media
             ).update_xaxes(categoryorder ='max descending').show()

Copyright and Getting the Data

Copyright 2021 Abbrivia GmbH https://www.abbrivia.com CC-BY (By Attribution) 4.0 https://creativecommons.org/licenses/by/4.0/legalcode Reuse our work freely!

All visualizations, and code produced in this notebook are completely open access under the Creative Commons BY license. You have the permission to use, distribute, and reproduce these in any medium, provided the source and authors are credited.

The data produced by third parties and made available by "Our World in Data" is subject to the license terms from the original third-party authors. Check the license of any third-party data before use and redistribution on 'https://ourworldindata.org/coronavirus' site (see below).

See the defintions and further discussion on the used dataset at the "Our World in Data" site https://ourworldindata.org/covid-vaccinations

The data is taken specifically from https://covid.ourworldindata.org/data/owid-covid-data.xlsx file

Hannah Ritchie, Edouard Mathieu, Lucas Rodés-Guirao, Cameron Appel, Charlie Giattino, Esteban Ortiz-Ospina, Joe Hasell, Bobbie Macdonald, Diana Beltekian and Max Roser (2020) - "Coronavirus Pandemic (COVID-19)". Published online at OurWorldInData.org. Retrieved from: 'https://ourworldindata.org/coronavirus' [Online Resource]

We use Excel file because it contains the data format information in itself. If you want to run this more often consider manually downloading the data and sourcing it locally as shown in the next line (commented out).